The purpose of this study was to clarify whether a gait analysis using an accelerometer could estimate gait independence. Eighty-six stroke patients and 21 healthy control subjects participated in this study. Stroke patients were identified as dependent or independent based on their gait ability. The acceleration of the trunk and bilateral thigh was measured using three wireless sensors during walking. The root mean square, gait regularity, and symmetry were calculated from the acceleration to estimate gait quality. ANCOVA showed that gait regularity of the trunk and bilateral thigh were significantly lowest in the dependent group, regardless of gait velocity. A logistic regression analysis showed that the regularity and root mean square of the anteroposterior acceleration of the unaffected thigh were the key factors for estimating gait independence. This study suggests that an acceleration-based gait analysis facilities gait independence estimation, and is a useful tool during the rehabilitation of stroke patients.